class Exhibitor:
def __init__(self, exhibitor, chroot):
self.chroot = chroot
self.exhibitor = ExhibitorEnsembleProvider(exhibitor['hosts'], exhibitor['port'], poll_interval=30)
self.client = KazooClient(hosts=self.exhibitor.zookeeper_hosts + self.chroot,
command_retry={
'deadline': 10,
'max_delay': 1,
'max_tries': -1},
connection_retry={'max_delay': 1, 'max_tries': -1})
self.client.add_listener(self.session_listener)
self.client.start()
def session_listener(self, state):
pass
def _poll_exhibitor(self):
if self.exhibitor.poll():
self.client.set_hosts(self.exhibitor.zookeeper_hosts + self.chroot)
def get(self, *params):
self._poll_exhibitor()
return self.client.retry(self.client.get, *params)
def get_children(self, *params):
self._poll_exhibitor()
try:
return self.client.retry(self.client.get_children, *params)
except NoNodeError:
return []
def test_update_host_list(self):
from kazoo.client import KazooClient
from kazoo.protocol.states import KeeperState
hosts = self.cluster[0].address
# create a client with only one server in its list
client = KazooClient(hosts=hosts)
client.start()
# try to change the chroot, not currently allowed
self.assertRaises(ConfigurationError,
client.set_hosts, hosts + '/new_chroot')
# grow the cluster to 3
client.set_hosts(self.servers)
# shut down the first host
try:
self.cluster[0].stop()
time.sleep(5)
eq_(client.client_state, KeeperState.CONNECTED)
finally:
self.cluster[0].run()
def test_update_host_list(self):
from kazoo.client import KazooClient
from kazoo.protocol.states import KeeperState
hosts = self.cluster[0].address
# create a client with only one server in its list
client = KazooClient(hosts=hosts)
client.start()
# try to change the chroot, not currently allowed
self.assertRaises(ConfigurationError, client.set_hosts,
hosts + '/new_chroot')
# grow the cluster to 3
client.set_hosts(self.servers)
# shut down the first host
try:
self.cluster[0].stop()
time.sleep(5)
eq_(client.client_state, KeeperState.CONNECTED)
finally:
self.cluster[0].run()
class ZooKeeper(object):
'''
Class implementing the ZooKeeper interface.
This class uses the facade design pattern to keep common interaction
with the ZooKeeper API simple and consistent for the caller, and
limits coupling between objects. It allows for more complex interactions
by providing direct access to the client connection when needed (though
that is discouraged). It also provides for a convenient entry point for
testing only ZooKeeper interactions.
'''
log = logging.getLogger("zuul.zk.ZooKeeper")
REQUEST_ROOT = '/nodepool/requests'
NODE_ROOT = '/nodepool/nodes'
# Log zookeeper retry every 10 seconds
retry_log_rate = 10
def __init__(self):
'''
Initialize the ZooKeeper object.
'''
self.client = None
self._became_lost = False
self._last_retry_log = 0
def _dictToStr(self, data):
return json.dumps(data).encode('utf8')
def _strToDict(self, data):
return json.loads(data.decode('utf8'))
def _connection_listener(self, state):
'''
Listener method for Kazoo connection state changes.
.. warning:: This method must not block.
'''
if state == KazooState.LOST:
self.log.debug("ZooKeeper connection: LOST")
self._became_lost = True
elif state == KazooState.SUSPENDED:
self.log.debug("ZooKeeper connection: SUSPENDED")
else:
self.log.debug("ZooKeeper connection: CONNECTED")
@property
def connected(self):
return self.client.state == KazooState.CONNECTED
@property
def suspended(self):
return self.client.state == KazooState.SUSPENDED
@property
def lost(self):
return self.client.state == KazooState.LOST
@property
def didLoseConnection(self):
return self._became_lost
def resetLostFlag(self):
self._became_lost = False
def logConnectionRetryEvent(self):
now = time.monotonic()
if now - self._last_retry_log >= self.retry_log_rate:
self.log.warning("Retrying zookeeper connection")
self._last_retry_log = now
def connect(self, hosts, read_only=False, timeout=10.0):
'''
Establish a connection with ZooKeeper cluster.
Convenience method if a pre-existing ZooKeeper connection is not
supplied to the ZooKeeper object at instantiation time.
:param str hosts: Comma-separated list of hosts to connect to (e.g.
127.0.0.1:2181,127.0.0.1:2182,[::1]:2183).
:param bool read_only: If True, establishes a read-only connection.
:param float timeout: The ZooKeeper session timeout, in
seconds (default: 10.0).
'''
if self.client is None:
self.client = KazooClient(hosts=hosts,
read_only=read_only,
timeout=timeout)
self.client.add_listener(self._connection_listener)
# Manually retry initial connection attempt
while True:
try:
self.client.start(1)
break
except KazooTimeoutError:
self.logConnectionRetryEvent()
def disconnect(self):
'''
Close the ZooKeeper cluster connection.
You should call this method if you used connect() to establish a
cluster connection.
'''
if self.client is not None and self.client.connected:
self.client.stop()
self.client.close()
self.client = None
def resetHosts(self, hosts):
'''
Reset the ZooKeeper cluster connection host list.
:param str hosts: Comma-separated list of hosts to connect to (e.g.
127.0.0.1:2181,127.0.0.1:2182,[::1]:2183).
'''
if self.client is not None:
self.client.set_hosts(hosts=hosts)
def submitNodeRequest(self, node_request, watcher):
'''
Submit a request for nodes to Nodepool.
:param NodeRequest node_request: A NodeRequest with the
contents of the request.
:param callable watcher: A callable object that will be
invoked each time the request is updated. It is called
with two arguments: (node_request, deleted) where
node_request is the same argument passed to this method,
and deleted is a boolean which is True if the node no
longer exists (notably, this will happen on disconnection
from ZooKeeper). The watcher should return False when
further updates are no longer necessary.
'''
data = node_request.toDict()
data['created_time'] = time.time()
path = '%s/%s-' % (self.REQUEST_ROOT, node_request.priority)
path = self.client.create(path,
self._dictToStr(data),
makepath=True,
sequence=True,
ephemeral=True)
reqid = path.split("/")[-1]
node_request.id = reqid
def callback(data, stat):
if data:
data = self._strToDict(data)
request_nodes = list(node_request.nodeset.getNodes())
for i, nodeid in enumerate(data.get('nodes', [])):
node_path = '%s/%s' % (self.NODE_ROOT, nodeid)
node_data, node_stat = self.client.get(node_path)
node_data = self._strToDict(node_data)
request_nodes[i].id = nodeid
request_nodes[i].updateFromDict(node_data)
node_request.updateFromDict(data)
deleted = (data is None) # data *are* none
return watcher(node_request, deleted)
self.client.DataWatch(path, callback)
def deleteNodeRequest(self, node_request):
'''
Delete a request for nodes.
:param NodeRequest node_request: A NodeRequest with the
contents of the request.
'''
path = '%s/%s' % (self.REQUEST_ROOT, node_request.id)
try:
self.client.delete(path)
except kze.NoNodeError:
pass
def nodeRequestExists(self, node_request):
'''
See if a NodeRequest exists in ZooKeeper.
:param NodeRequest node_request: A NodeRequest to verify.
:returns: True if the request exists, False otherwise.
'''
path = '%s/%s' % (self.REQUEST_ROOT, node_request.id)
if self.client.exists(path):
return True
return False
def storeNode(self, node):
'''Store the node.
The node is expected to already exist and is updated in its
entirety.
:param Node node: The node to update.
'''
path = '%s/%s' % (self.NODE_ROOT, node.id)
self.client.set(path, self._dictToStr(node.toDict()))
def lockNode(self, node, blocking=True, timeout=None):
'''
Lock a node.
This should be called as soon as a request is fulfilled and
the lock held for as long as the node is in-use. It can be
used by nodepool to detect if Zuul has gone offline and the
node should be reclaimed.
:param Node node: The node which should be locked.
'''
lock_path = '%s/%s/lock' % (self.NODE_ROOT, node.id)
try:
lock = Lock(self.client, lock_path)
have_lock = lock.acquire(blocking, timeout)
except kze.LockTimeout:
raise LockException("Timeout trying to acquire lock %s" %
lock_path)
# If we aren't blocking, it's possible we didn't get the lock
# because someone else has it.
if not have_lock:
raise LockException("Did not get lock on %s" % lock_path)
node.lock = lock
def unlockNode(self, node):
'''
Unlock a node.
The node must already have been locked.
:param Node node: The node which should be unlocked.
'''
if node.lock is None:
raise LockException("Node %s does not hold a lock" % (node, ))
node.lock.release()
node.lock = None
def heldNodeCount(self, autohold_key):
'''
Count the number of nodes being held for the given tenant/project/job.
:param set autohold_key: A set with the tenant/project/job names.
'''
identifier = " ".join(autohold_key)
try:
nodes = self.client.get_children(self.NODE_ROOT)
except kze.NoNodeError:
return 0
count = 0
for nodeid in nodes:
node_path = '%s/%s' % (self.NODE_ROOT, nodeid)
node_data, node_stat = self.client.get(node_path)
if not node_data:
self.log.warning("Node ID %s has no data", nodeid)
continue
node_data = self._strToDict(node_data)
if (node_data['state'] == zuul.model.STATE_HOLD
and node_data.get('hold_job') == identifier):
count += 1
return count
class ZooKeeper(AbstractDCS):
def __init__(self, name, config):
super(ZooKeeper, self).__init__(name, config)
hosts = config.get('hosts', [])
if isinstance(hosts, list):
hosts = ','.join(hosts)
self.exhibitor = None
if 'exhibitor' in config:
exhibitor = config['exhibitor']
interval = exhibitor.get('poll_interval', 300)
self.exhibitor = ExhibitorEnsembleProvider(exhibitor['hosts'], exhibitor['port'], poll_interval=interval)
hosts = self.exhibitor.zookeeper_hosts
self._client = KazooClient(hosts=hosts, timeout=(config.get('session_timeout') or 30),
command_retry={'deadline': (config.get('reconnect_timeout') or 10),
'max_delay': 1, 'max_tries': -1},
connection_retry={'max_delay': 1, 'max_tries': -1})
self._client.add_listener(self.session_listener)
self._my_member_data = None
self._fetch_cluster = True
self._last_leader_operation = 0
self._client.start()
def session_listener(self, state):
if state in [KazooState.SUSPENDED, KazooState.LOST]:
self.cluster_watcher(None)
def cluster_watcher(self, event):
self._fetch_cluster = True
self.event.set()
def get_node(self, key, watch=None):
try:
ret = self._client.get(key, watch)
return (ret[0].decode('utf-8'), ret[1])
except NoNodeError:
return None
@staticmethod
def member(name, value, znode):
return Member.from_node(znode.version, name, znode.ephemeralOwner, value)
def get_children(self, key, watch=None):
try:
return self._client.get_children(key, watch)
except NoNodeError:
return []
def load_members(self):
members = []
for member in self.get_children(self.members_path, self.cluster_watcher):
data = self.get_node(self.members_path + member)
if data is not None:
members.append(self.member(member, *data))
return members
def _inner_load_cluster(self):
self._fetch_cluster = False
self.event.clear()
nodes = set(self.get_children(self.client_path(''), self.cluster_watcher))
if not nodes:
self._fetch_cluster = True
# get initialize flag
initialize = (self.get_node(self.initialize_path) or [None])[0] if self._INITIALIZE in nodes else None
# get list of members
members = self.load_members() if self._MEMBERS[:-1] in nodes else []
# get leader
leader = self.get_node(self.leader_path) if self._LEADER in nodes else None
if leader:
client_id = self._client.client_id
if leader[0] == self._name and client_id is not None and client_id[0] != leader[1].ephemeralOwner:
logger.info('I am leader but not owner of the session. Removing leader node')
self._client.delete(self.leader_path)
leader = None
if leader:
member = Member(-1, leader[0], None, {})
member = ([m for m in members if m.name == leader[0]] or [member])[0]
leader = Leader(leader[1].version, leader[1].ephemeralOwner, member)
self._fetch_cluster = member.index == -1
# failover key
failover = self.get_node(self.failover_path, watch=self.cluster_watcher) if self._FAILOVER in nodes else None
if failover:
failover = Failover.from_node(failover[1].version, failover[0])
# get last leader operation
optime = self.get_node(self.leader_optime_path) if self._OPTIME in nodes and self._fetch_cluster else None
self._last_leader_operation = 0 if optime is None else int(optime[0])
self._cluster = Cluster(initialize, leader, self._last_leader_operation, members, failover)
def _load_cluster(self):
if self.exhibitor and self.exhibitor.poll():
self._client.set_hosts(self.exhibitor.zookeeper_hosts)
if self._fetch_cluster or self._cluster is None:
try:
self._client.retry(self._inner_load_cluster)
except:
logger.exception('get_cluster')
self.session_listener(KazooState.LOST)
raise ZooKeeperError('ZooKeeper in not responding properly')
def _create(self, path, value, **kwargs):
try:
self._client.retry(self._client.create, path, value.encode('utf-8'), **kwargs)
return True
except:
return False
def attempt_to_acquire_leader(self):
ret = self._create(self.leader_path, self._name, makepath=True, ephemeral=True)
if not ret:
logger.info('Could not take out TTL lock')
return ret
def set_failover_value(self, value, index=None):
try:
self._client.retry(self._client.set, self.failover_path, value.encode('utf-8'), version=index or -1)
return True
except NoNodeError:
return value == '' or (not index and self._create(self.failover_path, value))
except:
logging.exception('set_failover_value')
return False
def initialize(self, create_new=True, sysid=""):
return self._create(self.initialize_path, sysid, makepath=True) if create_new \
else self._client.retry(self._client.set, self.initialize_path, sysid.encode("utf-8"))
def touch_member(self, data, ttl=None):
cluster = self.cluster
member = cluster and ([m for m in cluster.members if m.name == self._name] or [None])[0]
path = self.member_path
data = data.encode('utf-8')
if member and self._client.client_id is not None and member.session != self._client.client_id[0]:
try:
self._client.retry(self._client.delete, path)
except NoNodeError:
pass
except:
return False
member = None
if member and data == self._my_member_data:
return True
try:
if member:
self._client.retry(self._client.set, path, data)
else:
self._client.retry(self._client.create, path, data, makepath=True, ephemeral=True)
self._my_member_data = data
return True
except NodeExistsError:
try:
self._client.retry(self._client.set, path, data)
self._my_member_data = data
return True
except:
logger.exception('touch_member')
except:
logger.exception('touch_member')
return False
def take_leader(self):
return self.attempt_to_acquire_leader()
def write_leader_optime(self, last_operation):
last_operation = last_operation.encode('utf-8')
if last_operation != self._last_leader_operation:
self._last_leader_operation = last_operation
path = self.leader_optime_path
try:
self._client.retry(self._client.set, path, last_operation)
except NoNodeError:
try:
self._client.retry(self._client.create, path, last_operation, makepath=True)
except:
logger.exception('Failed to create %s', path)
except:
logger.exception('Failed to update %s', path)
def update_leader(self):
return True
def delete_leader(self):
self._client.restart()
self._my_member_data = None
return True
def _cancel_initialization(self):
node = self.get_node(self.initialize_path)
if node:
self._client.delete(self.initialize_path, version=node[1].version)
def cancel_initialization(self):
try:
self._client.retry(self._cancel_initialization)
except:
logger.exception("Unable to delete initialize key")
def delete_cluster(self):
try:
return self._client.retry(self._client.delete, self.client_path(''), recursive=True)
except NoNodeError:
return True
def watch(self, timeout):
if super(ZooKeeper, self).watch(timeout):
self._fetch_cluster = True
return self._fetch_cluster
class ZooKeeper(object):
'''
Class implementing the ZooKeeper interface.
This class uses the facade design pattern to keep common interaction
with the ZooKeeper API simple and consistent for the caller, and
limits coupling between objects. It allows for more complex interactions
by providing direct access to the client connection when needed (though
that is discouraged). It also provides for a convenient entry point for
testing only ZooKeeper interactions.
Most API calls reference an image name only, as the path for the znode
for that image is calculated automatically. And image names are assumed
to be unique.
'''
log = logging.getLogger("nodepool.zk.ZooKeeper")
IMAGE_ROOT = "/nodepool/images"
def __init__(self):
'''
Initialize the ZooKeeper object.
'''
self.client = None
self._became_lost = False
#========================================================================
# Private Methods
#========================================================================
def _imagePath(self, image):
return "%s/%s" % (self.IMAGE_ROOT, image)
def _imageBuildRequestPath(self, image):
return "%s/request-build" % self._imagePath(image)
def _imageBuildsPath(self, image):
return "%s/builds" % self._imagePath(image)
def _imageBuildLockPath(self, image):
return "%s/lock" % self._imageBuildsPath(image)
def _imageBuildNumberLockPath(self, image, build_number):
return "%s/%s/lock" % (self._imageBuildsPath(image),
build_number)
def _imageProviderPath(self, image, build_number):
return "%s/%s/providers" % (self._imageBuildsPath(image),
build_number)
def _imageUploadPath(self, image, build_number, provider):
return "%s/%s/providers/%s/images" % (self._imageBuildsPath(image),
build_number,
provider)
def _imageUploadLockPath(self, image, build_number, provider):
return "%s/lock" % self._imageUploadPath(image, build_number,
provider)
def _dictToStr(self, data):
return json.dumps(data)
def _strToDict(self, data):
return json.loads(data)
def _getImageBuildLock(self, image, blocking=True, timeout=None):
lock_path = self._imageBuildLockPath(image)
try:
lock = Lock(self.client, lock_path)
have_lock = lock.acquire(blocking, timeout)
except kze.LockTimeout:
raise npe.TimeoutException(
"Timeout trying to acquire lock %s" % lock_path)
# If we aren't blocking, it's possible we didn't get the lock
# because someone else has it.
if not have_lock:
raise npe.ZKLockException("Did not get lock on %s" % lock_path)
return lock
def _getImageBuildNumberLock(self, image, build_number,
blocking=True, timeout=None):
lock_path = self._imageBuildNumberLockPath(image, build_number)
try:
lock = Lock(self.client, lock_path)
have_lock = lock.acquire(blocking, timeout)
except kze.LockTimeout:
raise npe.TimeoutException(
"Timeout trying to acquire lock %s" % lock_path)
# If we aren't blocking, it's possible we didn't get the lock
# because someone else has it.
if not have_lock:
raise npe.ZKLockException("Did not get lock on %s" % lock_path)
return lock
def _getImageUploadLock(self, image, build_number, provider,
blocking=True, timeout=None):
lock_path = self._imageUploadLockPath(image, build_number, provider)
try:
lock = Lock(self.client, lock_path)
have_lock = lock.acquire(blocking, timeout)
except kze.LockTimeout:
raise npe.TimeoutException(
"Timeout trying to acquire lock %s" % lock_path)
# If we aren't blocking, it's possible we didn't get the lock
# because someone else has it.
if not have_lock:
raise npe.ZKLockException("Did not get lock on %s" % lock_path)
return lock
def _connection_listener(self, state):
'''
Listener method for Kazoo connection state changes.
.. warning:: This method must not block.
'''
if state == KazooState.LOST:
self.log.debug("ZooKeeper connection: LOST")
self._became_lost = True
elif state == KazooState.SUSPENDED:
self.log.debug("ZooKeeper connection: SUSPENDED")
else:
self.log.debug("ZooKeeper connection: CONNECTED")
#========================================================================
# Public Methods and Properties
#========================================================================
@property
def connected(self):
return self.client.state == KazooState.CONNECTED
@property
def suspended(self):
return self.client.state == KazooState.SUSPENDED
@property
def lost(self):
return self.client.state == KazooState.LOST
@property
def didLoseConnection(self):
return self._became_lost
def resetLostFlag(self):
self._became_lost = False
def connect(self, host_list, read_only=False):
'''
Establish a connection with ZooKeeper cluster.
Convenience method if a pre-existing ZooKeeper connection is not
supplied to the ZooKeeper object at instantiation time.
:param list host_list: A list of
:py:class:`~nodepool.zk.ZooKeeperConnectionConfig` objects
(one per server) defining the ZooKeeper cluster servers.
:param bool read_only: If True, establishes a read-only connection.
'''
if self.client is None:
hosts = buildZooKeeperHosts(host_list)
self.client = KazooClient(hosts=hosts, read_only=read_only)
self.client.add_listener(self._connection_listener)
self.client.start()
def disconnect(self):
'''
Close the ZooKeeper cluster connection.
You should call this method if you used connect() to establish a
cluster connection.
'''
if self.client is not None and self.client.connected:
self.client.stop()
self.client.close()
self.client = None
def resetHosts(self, host_list):
'''
Reset the ZooKeeper cluster connection host list.
:param list host_list: A list of
:py:class:`~nodepool.zk.ZooKeeperConnectionConfig` objects
(one per server) defining the ZooKeeper cluster servers.
'''
if self.client is not None:
hosts = buildZooKeeperHosts(host_list)
self.client.set_hosts(hosts=hosts)
@contextmanager
def imageBuildLock(self, image, blocking=True, timeout=None):
'''
Context manager to use for locking image builds.
Obtains a write lock for the specified image.
:param str image: Name of the image to lock
:param bool blocking: Whether or not to block on trying to
acquire the lock
:param int timeout: When blocking, how long to wait for the lock
to get acquired. None, the default, waits forever.
:raises: TimeoutException if we failed to acquire the lock when
blocking with a timeout. ZKLockException if we are not blocking
and could not get the lock, or a lock is already held.
'''
lock = None
try:
lock = self._getImageBuildLock(image, blocking, timeout)
yield
finally:
if lock:
lock.release()
@contextmanager
def imageBuildNumberLock(self, image, build_number,
blocking=True, timeout=None):
'''
Context manager to use for locking _specific_ image builds.
Obtains a write lock for the specified image build number. This is
used for locking a build number during the cleanup phase of the
builder.
:param str image: Name of the image
:param str build_number: The image build number to lock.
:param bool blocking: Whether or not to block on trying to
acquire the lock
:param int timeout: When blocking, how long to wait for the lock
to get acquired. None, the default, waits forever.
:raises: TimeoutException if we failed to acquire the lock when
blocking with a timeout. ZKLockException if we are not blocking
and could not get the lock, or a lock is already held.
'''
lock = None
try:
lock = self._getImageBuildNumberLock(image, build_number,
blocking, timeout)
yield
finally:
if lock:
lock.release()
@contextmanager
def imageUploadLock(self, image, build_number, provider,
blocking=True, timeout=None):
'''
Context manager to use for locking image builds.
Obtains a write lock for the specified image upload.
:param str image: Name of the image.
:param str build_number: The image build number.
:param str provider: The provider name owning the image.
:param bool blocking: Whether or not to block on trying to
acquire the lock
:param int timeout: When blocking, how long to wait for the lock
to get acquired. None, the default, waits forever.
:raises: TimeoutException if we failed to acquire the lock when
blocking with a timeout. ZKLockException if we are not blocking
and could not get the lock, or a lock is already held.
'''
lock = None
try:
lock = self._getImageUploadLock(image, build_number, provider,
blocking, timeout)
yield
finally:
if lock:
lock.release()
def getImageNames(self):
'''
Retrieve the image names in Zookeeper.
:returns: A list of image names or the empty list.
'''
path = self.IMAGE_ROOT
try:
images = self.client.get_children(path)
except kze.NoNodeError:
return []
return sorted(images)
def getBuildNumbers(self, image):
'''
Retrieve the builds available for an image.
:param str image: The image name.
:returns: A list of image build numbers or the empty list.
'''
path = self._imageBuildsPath(image)
try:
builds = self.client.get_children(path)
except kze.NoNodeError:
return []
builds = [x for x in builds if x != 'lock']
return builds
def getBuildProviders(self, image, build_number):
'''
Retrieve the providers which have uploads for an image build.
:param str image: The image name.
:param str build_number: The image build number.
:returns: A list of provider names or the empty list.
'''
path = self._imageProviderPath(image, build_number)
try:
providers = self.client.get_children(path)
except kze.NoNodeError:
return []
return sorted(providers)
def getImageUploadNumbers(self, image, build_number, provider):
'''
Retrieve upload numbers for a provider and image build.
:param str image: The image name.
:param str build_number: The image build number.
:param str provider: The provider name owning the image.
:returns: A list of upload numbers or the empty list.
'''
path = self._imageUploadPath(image, build_number, provider)
try:
uploads = self.client.get_children(path)
except kze.NoNodeError:
return []
uploads = [x for x in uploads if x != 'lock']
return uploads
def getBuild(self, image, build_number):
'''
Retrieve the image build data.
:param str image: The image name.
:param str build_number: The image build number.
:returns: An ImageBuild object, or None if not found.
'''
path = self._imageBuildsPath(image) + "/%s" % build_number
try:
data, stat = self.client.get(path)
except kze.NoNodeError:
return None
d = ImageBuild.fromDict(self._strToDict(data), build_number)
d.stat = stat
return d
def getBuilds(self, image, states=None):
'''
Retrieve all image build data matching any given states.
:param str image: The image name.
:param list states: A list of build state values to match against.
A value of None will disable state matching and just return
all builds.
:returns: A list of ImageBuild objects.
'''
path = self._imageBuildsPath(image)
try:
builds = self.client.get_children(path)
except kze.NoNodeError:
return []
matches = []
for build in builds:
if build == 'lock': # skip the build lock node
continue
data = self.getBuild(image, build)
if states is None:
matches.append(data)
elif data and data.state in states:
matches.append(data)
return matches
def getMostRecentBuilds(self, count, image, state=None):
'''
Retrieve the most recent image build data with the given state.
:param int count: A count of the most recent builds to return.
Use None for all builds.
:param str image: The image name.
:param str state: The build state to match on. Use None to
ignore state
:returns: A list of the most recent ImageBuild objects matching the
given state, or an empty list if there were no builds matching
the state. You may get less than 'count' entries if there were
not enough matching builds.
'''
states = None
if state:
states = [state]
builds = self.getBuilds(image, states)
if not builds:
return []
builds.sort(key=lambda x: x.state_time, reverse=True)
return builds[:count]
def storeBuild(self, image, build_data, build_number=None):
'''
Store the image build data.
The build data is either created if it does not exist, or it is
updated in its entirety if it does not. There is no partial updating.
The build data is expected to be represented as a dict. This dict may
contain any data, as appropriate.
If a build number is not supplied, then a new build node/number is
created. The new build number is available in the return value.
.. important: You should have the image locked before calling this
method.
:param str image: The image name for which we have data.
:param ImageBuild build_data: The build data.
:param str build_number: The image build number.
:returns: A string for the build number that was updated.
'''
# Append trailing / so the sequence node is created as a child node.
build_path = self._imageBuildsPath(image) + "/"
if build_number is None:
path = self.client.create(
build_path,
value=self._dictToStr(build_data.toDict()),
sequence=True,
makepath=True)
build_number = path.split("/")[-1]
else:
path = build_path + build_number
self.client.set(path, self._dictToStr(build_data.toDict()))
return build_number
def getImageUpload(self, image, build_number, provider, upload_number):
'''
Retrieve the image upload data.
:param str image: The image name.
:param str build_number: The image build number.
:param str provider: The provider name owning the image.
:param str upload_number: The image upload number.
:returns: An ImageUpload object, or None if not found.
:raises: ZKException if the image upload path is not found.
'''
path = self._imageUploadPath(image, build_number, provider)
path = path + "/%s" % upload_number
try:
data, stat = self.client.get(path)
except kze.NoNodeError:
return None
d = ImageUpload.fromDict(
self._strToDict(data), build_number, provider, image, upload_number
)
d.stat = stat
return d
def getUploads(self, image, build_number, provider, states=None):
'''
Retrieve all image upload data matching any given states.
:param str image: The image name.
:param str build_number: The image build number.
:param str provider: The provider name owning the image.
:param list states: A list of upload state values to match against.
A value of None will disable state matching and just return
all uploads.
:returns: A list of ImageUpload objects.
'''
path = self._imageUploadPath(image, build_number, provider)
try:
uploads = self.client.get_children(path)
except kze.NoNodeError:
return []
matches = []
for upload in uploads:
if upload == 'lock':
continue
data = self.getImageUpload(image, build_number, provider, upload)
if states is None:
matches.append(data)
elif data and data.state in states:
matches.append(data)
return matches
def getMostRecentBuildImageUploads(self, count, image, build_number,
provider, state=None):
'''
Retrieve the most recent image upload data with the given state.
:param int count: A count of the most recent uploads to return.
Use None for all uploads.
:param str image: The image name.
:param str build_number: The image build number.
:param str provider: The provider name owning the image.
:param str state: The image upload state to match on. Use None to
ignore state.
:returns: A tuple with the most recent upload number and dictionary of
upload data matching the given state, or None if there was no
upload matching the state.
'''
states = None
if state:
states = [state]
uploads = self.getUploads(image, build_number, provider, states)
if not uploads:
return []
uploads.sort(key=lambda x: x.state_time, reverse=True)
return uploads[:count]
def getMostRecentImageUpload(self, image, provider,
state="ready"):
'''
Retrieve the most recent image upload data with the given state.
:param str image: The image name.
:param str provider: The provider name owning the image.
:param str state: The image upload state to match on.
:returns: An ImageUpload object matching the given state, or
None if there is no recent upload.
'''
recent_data = None
for build_number in self.getBuildNumbers(image):
path = self._imageUploadPath(image, build_number, provider)
try:
uploads = self.client.get_children(path)
except kze.NoNodeError:
uploads = []
for upload in uploads:
if upload == 'lock': # skip the upload lock node
continue
data = self.getImageUpload(image, build_number, provider, upload)
if not data or data.state != state:
continue
elif (recent_data is None or
recent_data.state_time < data.state_time):
recent_data = data
return recent_data
def storeImageUpload(self, image, build_number, provider, image_data,
upload_number=None):
'''
Store the built image's upload data for the given provider.
The image upload data is either created if it does not exist, or it
is updated in its entirety if it does not. There is no partial
updating. The image data is expected to be represented as a dict.
This dict may contain any data, as appropriate.
If an image upload number is not supplied, then a new image upload
node/number is created. The new upload number is available in the
return value.
:param str image: The image name for which we have data.
:param str build_number: The image build number.
:param str provider: The provider name owning the image.
:param ImageUpload image_data: The image data we want to store.
:param str upload_number: The image upload number to update.
:returns: A string for the upload number that was updated.
:raises: ZKException for an invalid image build.
'''
# We expect the image builds path to already exist.
build_path = self._imageBuildsPath(image)
if not self.client.exists(build_path):
raise npe.ZKException(
"Cannot find build %s of image %s" % (build_number, image)
)
# Generate a path for the upload. This doesn't have to exist yet
# since we'll create new provider/upload ID znodes automatically.
# Append trailing / so the sequence node is created as a child node.
upload_path = self._imageUploadPath(image, build_number, provider) + "/"
if upload_number is None:
path = self.client.create(
upload_path,
value=self._dictToStr(image_data.toDict()),
sequence=True,
makepath=True)
upload_number = path.split("/")[-1]
else:
path = upload_path + upload_number
self.client.set(path, self._dictToStr(image_data.toDict()))
return upload_number
def hasBuildRequest(self, image):
'''
Check if an image has a pending build request.
:param str image: The image name to check.
:returns: True if request is pending, False otherwise
'''
path = self._imageBuildRequestPath(image)
if self.client.exists(path) is not None:
return True
return False
def submitBuildRequest(self, image):
'''
Submit a request for a new image build.
:param str image: The image name.
'''
path = self._imageBuildRequestPath(image)
self.client.ensure_path(path)
def removeBuildRequest(self, image):
'''
Remove an image build request.
:param str image: The image name to check.
'''
path = self._imageBuildRequestPath(image)
try:
self.client.delete(path)
except kze.NoNodeError:
pass
def deleteBuild(self, image, build_number):
'''
Delete an image build from ZooKeeper.
Any provider uploads for this build must be deleted before the build
node can be deleted.
:param str image: The image name.
:param str build_number: The image build number to delete.
:returns: True if the build is successfully deleted or did not exist,
False if the provider uploads still exist.
'''
path = self._imageBuildsPath(image)
path = path + "/%s" % build_number
# Verify that no upload znodes exist.
for prov in self.getBuildProviders(image, build_number):
if self.getImageUploadNumbers(image, build_number, prov):
return False
try:
# NOTE: Need to do recursively to remove lock znodes
self.client.delete(path, recursive=True)
except kze.NoNodeError:
pass
return True
def deleteUpload(self, image, build_number, provider, upload_number):
'''
Delete an image upload from ZooKeeper.
:param str image: The image name.
:param str build_number: The image build number.
:param str provider: The provider name owning the image.
:param str upload_number: The image upload number to delete.
'''
path = self._imageUploadPath(image, build_number, provider)
path = path + "/%s" % upload_number
try:
self.client.delete(path)
except kze.NoNodeError:
pass
class _ZookeeperProxy(object):
def __init__(self, address_provider: AddressListProvider, prefix: str):
self.address_provider = address_provider
self.async_counter = WaitingCounter(limit=100)
self.conn_str = None
self.client = None
self.prefix = prefix
self.hosts_cache = SlowlyUpdatedCache(
self.address_provider.get_latest_address,
self._update_hosts,
30, # Refresh every 30 seconds
3 * 60) # Update only after 180 seconds of stability
def _update_hosts(self, value):
hosts, port = value
if hosts:
self.conn_str = ','.join(['{}:{}'.format(h, port)
for h in hosts]) + self.prefix
if self.client is None:
self.client = KazooClient(hosts=self.conn_str,
command_retry={
'deadline': 120,
'max_delay': 1,
'max_tries': -1
},
connection_retry={
'max_delay': 1,
'max_tries': -1
})
self.client.add_listener(self.session_listener)
else:
self.client.stop()
self.client.set_hosts(self.conn_str)
self.client.start()
def terminate(self):
if self.client:
self.client.stop()
def session_listener(self, state):
pass
def get_conn_str(self):
return self.conn_str
def get(self, *params):
self.hosts_cache.touch()
return self.client.retry(self.client.get, *params)
def get_async(self, *params):
# Exhibitor is not polled here and it's totally fine!
self.async_counter.increment()
try:
i_async = self.client.get_async(*params)
i_async.rawlink(self._decrement)
return i_async
except Exception as e:
self._decrement()
raise e
def _decrement(self, *args, **kwargs):
self.async_counter.decrement()
def set(self, *args, **kwargs):
self.hosts_cache.touch()
return self.client.retry(self.client.set, *args, **kwargs)
def create(self, *args, **kwargs):
self.hosts_cache.touch()
return self.client.retry(self.client.create, *args, **kwargs)
def delete(self, *args, **kwargs):
self.hosts_cache.touch()
try:
return self.client.retry(self.client.delete, *args, **kwargs)
except NoNodeError:
pass
def get_children(self, *params):
self.hosts_cache.touch()
try:
return self.client.retry(self.client.get_children, *params)
except NoNodeError:
return []
def take_lock(self, *args, **kwargs):
while True:
try:
self.hosts_cache.touch()
return self.client.Lock(*args, **kwargs)
except Exception as e:
_LOG.error('Failed to obtain lock for exhibitor, retrying',
exc_info=e)
class ZooKeeper(AbstractDCS):
def __init__(self, name, config):
super(ZooKeeper, self).__init__(name, config)
hosts = config.get('hosts', [])
if isinstance(hosts, list):
hosts = ','.join(hosts)
self.exhibitor = None
if 'exhibitor' in config:
exhibitor = config['exhibitor']
interval = exhibitor.get('poll_interval', 300)
self.exhibitor = ExhibitorEnsembleProvider(exhibitor['hosts'], exhibitor['port'], poll_interval=interval)
hosts = self.exhibitor.zookeeper_hosts
self.client = KazooClient(hosts=hosts,
timeout=(config.get('session_timeout', None) or 30),
command_retry={
'deadline': (config.get('reconnect_timeout', None) or 10),
'max_delay': 1,
'max_tries': -1},
connection_retry={'max_delay': 1, 'max_tries': -1})
self.client.add_listener(self.session_listener)
self.cluster_event = self.client.handler.event_object()
self.cluster = None
self.fetch_cluster = True
self.last_leader_operation = 0
self.client.start(None)
def session_listener(self, state):
if state in [KazooState.SUSPENDED, KazooState.LOST]:
self.cluster_watcher(None)
def cluster_watcher(self, event):
self.fetch_cluster = True
self.cluster_event.set()
def get_node(self, key, watch=None):
try:
ret = self.client.get(key, watch)
return (ret[0].decode('utf-8'), ret[1])
except NoNodeError:
return None
@staticmethod
def member(name, value, znode):
conn_url, api_url = parse_connection_string(value)
return Member(znode.version, name, conn_url, api_url, None, None)
def get_children(self, key, watch=None):
try:
return self.client.get_children(key, watch)
except NoNodeError:
return []
def load_members(self):
members = []
for member in self.get_children(self.members_path, self.cluster_watcher):
data = self.get_node(self.members_path + member)
if data is not None:
members.append(self.member(member, *data))
return members
def _inner_load_cluster(self):
self.cluster_event.clear()
nodes = set(self.get_children(self.client_path('')))
# get initialize flag
initialize = self._INITIALIZE in nodes
# get list of members
members = self.load_members() if self._MEMBERS[:-1] in nodes else []
# get leader
leader = self.get_node(self.leader_path, self.cluster_watcher) if self._LEADER in nodes else None
if leader:
client_id = self.client.client_id
if leader[0] == self._name and client_id is not None and client_id[0] != leader[1].ephemeralOwner:
logger.info('I am leader but not owner of the session. Removing leader node')
self.client.delete(self.leader_path)
leader = None
if leader:
member = Member(-1, leader[0], None, None, None, None)
member = ([m for m in members if m.name == leader[0]] or [member])[0]
leader = Leader(leader[1].version, None, None, member)
self.fetch_cluster = member.index == -1
# get last leader operation
self.last_leader_operation = self.get_node(self.leader_optime_path) if self.fetch_cluster else None
self.last_leader_operation = 0 if self.last_leader_operation is None else int(self.last_leader_operation[0])
self.cluster = Cluster(initialize, leader, self.last_leader_operation, members)
def get_cluster(self):
if self.exhibitor and self.exhibitor.poll():
self.client.set_hosts(self.exhibitor.zookeeper_hosts)
if self.fetch_cluster:
try:
self.client.retry(self._inner_load_cluster)
except:
logger.exception('get_cluster')
self.session_listener(KazooState.LOST)
raise ZooKeeperError('ZooKeeper in not responding properly')
return self.cluster
def _create(self, path, value, **kwargs):
try:
self.client.retry(self.client.create, path, value.encode('utf-8'), **kwargs)
return True
except:
return False
def attempt_to_acquire_leader(self):
ret = self._create(self.leader_path, self._name, makepath=True, ephemeral=True)
ret or logger.info('Could not take out TTL lock')
return ret
def initialize(self):
return self._create(self.initialize_path, self._name, makepath=True)
def touch_member(self, connection_string, ttl=None):
if self.cluster and any(m.name == self._name for m in self.cluster.members):
return True
path = self.member_path
connection_string = connection_string.encode('utf-8')
try:
self.client.retry(self.client.create, path, connection_string, makepath=True, ephemeral=True)
return True
except NodeExistsError:
try:
self.client.retry(self.client.delete, path)
self.client.retry(self.client.create, path, connection_string, makepath=True, ephemeral=True)
return True
except:
logger.exception('touch_member')
return False
def take_leader(self):
return self.attempt_to_acquire_leader()
def update_leader(self, state_handler):
last_operation = state_handler.last_operation().encode('utf-8')
if last_operation != self.last_leader_operation:
self.last_leader_operation = last_operation
path = self.leader_optime_path
try:
self.client.retry(self.client.set, path, last_operation)
except NoNodeError:
try:
self.client.retry(self.client.create, path, last_operation, makepath=True)
except:
logger.exception('Failed to create %s', path)
except:
logger.exception('Failed to update %s', path)
return True
def delete_leader(self):
if isinstance(self.cluster, Cluster) and self.cluster.leader.name == self._name:
self.client.delete(self.leader_path, version=self.cluster.leader.index)
def _cancel_initialization(self):
node = self.get_node(self.initialize_path)
if node and node[0] == self._name:
self.client.delete(self.initialize_path, version=node[1].version)
def cancel_initialization(self):
try:
self.client.retry(self._cancel_initialization)
except:
logger.exception("Unable to delete initialize key")
def watch(self, timeout):
self.cluster_event.wait(timeout)
if self.cluster_event.isSet():
self.fetch_cluster = True
class ZooKeeper(object):
'''
Class implementing the ZooKeeper interface.
This class uses the facade design pattern to keep common interaction
with the ZooKeeper API simple and consistent for the caller, and
limits coupling between objects. It allows for more complex interactions
by providing direct access to the client connection when needed (though
that is discouraged). It also provides for a convenient entry point for
testing only ZooKeeper interactions.
'''
log = logging.getLogger("zuul.zk.ZooKeeper")
REQUEST_ROOT = '/nodepool/requests'
REQUEST_LOCK_ROOT = "/nodepool/requests-lock"
NODE_ROOT = '/nodepool/nodes'
HOLD_REQUEST_ROOT = '/zuul/hold-requests'
# Log zookeeper retry every 10 seconds
retry_log_rate = 10
def __init__(self, enable_cache=True):
'''
Initialize the ZooKeeper object.
:param bool enable_cache: When True, enables caching of ZooKeeper
objects (e.g., HoldRequests).
'''
self.client = None
self._became_lost = False
self._last_retry_log = 0
self.enable_cache = enable_cache
# The caching model we use is designed around handing out model
# data as objects. To do this, we use two caches: one is a TreeCache
# which contains raw znode data (among other details), and one for
# storing that data serialized as objects. This allows us to return
# objects from the APIs, and avoids calling the methods to serialize
# the data into objects more than once.
self._hold_request_tree = None
self._cached_hold_requests = {}
def _dictToStr(self, data):
return json.dumps(data).encode('utf8')
def _strToDict(self, data):
return json.loads(data.decode('utf8'))
def _connection_listener(self, state):
'''
Listener method for Kazoo connection state changes.
.. warning:: This method must not block.
'''
if state == KazooState.LOST:
self.log.debug("ZooKeeper connection: LOST")
self._became_lost = True
elif state == KazooState.SUSPENDED:
self.log.debug("ZooKeeper connection: SUSPENDED")
else:
self.log.debug("ZooKeeper connection: CONNECTED")
@property
def connected(self):
return self.client.state == KazooState.CONNECTED
@property
def suspended(self):
return self.client.state == KazooState.SUSPENDED
@property
def lost(self):
return self.client.state == KazooState.LOST
@property
def didLoseConnection(self):
return self._became_lost
def resetLostFlag(self):
self._became_lost = False
def logConnectionRetryEvent(self):
now = time.monotonic()
if now - self._last_retry_log >= self.retry_log_rate:
self.log.warning("Retrying zookeeper connection")
self._last_retry_log = now
def connect(self, hosts, read_only=False, timeout=10.0):
'''
Establish a connection with ZooKeeper cluster.
Convenience method if a pre-existing ZooKeeper connection is not
supplied to the ZooKeeper object at instantiation time.
:param str hosts: Comma-separated list of hosts to connect to (e.g.
127.0.0.1:2181,127.0.0.1:2182,[::1]:2183).
:param bool read_only: If True, establishes a read-only connection.
:param float timeout: The ZooKeeper session timeout, in
seconds (default: 10.0).
'''
if self.client is None:
self.client = KazooClient(hosts=hosts,
read_only=read_only,
timeout=timeout)
self.client.add_listener(self._connection_listener)
# Manually retry initial connection attempt
while True:
try:
self.client.start(1)
break
except KazooTimeoutError:
self.logConnectionRetryEvent()
if self.enable_cache:
self._hold_request_tree = TreeCache(self.client,
self.HOLD_REQUEST_ROOT)
self._hold_request_tree.listen_fault(self.cacheFaultListener)
self._hold_request_tree.listen(self.holdRequestCacheListener)
self._hold_request_tree.start()
def cacheFaultListener(self, e):
self.log.exception(e)
def holdRequestCacheListener(self, event):
'''
Keep the hold request object cache in sync with the TreeCache.
'''
try:
self._holdRequestCacheListener(event)
except Exception:
self.log.exception(
"Exception in hold request cache update for event: %s", event)
def _holdRequestCacheListener(self, event):
if hasattr(event.event_data, 'path'):
# Ignore root node
path = event.event_data.path
if path == self.HOLD_REQUEST_ROOT:
return
if event.event_type not in (TreeEvent.NODE_ADDED,
TreeEvent.NODE_UPDATED,
TreeEvent.NODE_REMOVED):
return
path = event.event_data.path
request_id = path.rsplit('/', 1)[1]
if event.event_type in (TreeEvent.NODE_ADDED, TreeEvent.NODE_UPDATED):
# Requests with no data are invalid
if not event.event_data.data:
return
# Perform an in-place update of the already cached request
d = self._bytesToDict(event.event_data.data)
old_request = self._cached_hold_requests.get(request_id)
if old_request:
if event.event_data.stat.version <= old_request.stat.version:
# Don't update to older data
return
old_request.updateFromDict(d)
old_request.stat = event.event_data.stat
else:
request = zuul.model.HoldRequest.fromDict(d)
request.id = request_id
request.stat = event.event_data.stat
self._cached_hold_requests[request_id] = request
elif event.event_type == TreeEvent.NODE_REMOVED:
try:
del self._cached_hold_requests[request_id]
except KeyError:
pass
def disconnect(self):
'''
Close the ZooKeeper cluster connection.
You should call this method if you used connect() to establish a
cluster connection.
'''
if self._hold_request_tree is not None:
self._hold_request_tree.close()
self._hold_request_tree = None
if self.client is not None and self.client.connected:
self.client.stop()
self.client.close()
self.client = None
def resetHosts(self, hosts):
'''
Reset the ZooKeeper cluster connection host list.
:param str hosts: Comma-separated list of hosts to connect to (e.g.
127.0.0.1:2181,127.0.0.1:2182,[::1]:2183).
'''
if self.client is not None:
self.client.set_hosts(hosts=hosts)
def submitNodeRequest(self, node_request, watcher):
'''
Submit a request for nodes to Nodepool.
:param NodeRequest node_request: A NodeRequest with the
contents of the request.
:param callable watcher: A callable object that will be
invoked each time the request is updated. It is called
with two arguments: (node_request, deleted) where
node_request is the same argument passed to this method,
and deleted is a boolean which is True if the node no
longer exists (notably, this will happen on disconnection
from ZooKeeper). The watcher should return False when
further updates are no longer necessary.
'''
node_request.created_time = time.time()
data = node_request.toDict()
path = '{}/{:0>3}-'.format(self.REQUEST_ROOT, node_request.priority)
path = self.client.create(path,
self._dictToStr(data),
makepath=True,
sequence=True,
ephemeral=True)
reqid = path.split("/")[-1]
node_request.id = reqid
def callback(data, stat):
if data:
self.updateNodeRequest(node_request, data)
deleted = (data is None) # data *are* none
return watcher(node_request, deleted)
self.client.DataWatch(path, callback)
def deleteNodeRequest(self, node_request):
'''
Delete a request for nodes.
:param NodeRequest node_request: A NodeRequest with the
contents of the request.
'''
path = '%s/%s' % (self.REQUEST_ROOT, node_request.id)
try:
self.client.delete(path)
except kze.NoNodeError:
pass
def nodeRequestExists(self, node_request):
'''
See if a NodeRequest exists in ZooKeeper.
:param NodeRequest node_request: A NodeRequest to verify.
:returns: True if the request exists, False otherwise.
'''
path = '%s/%s' % (self.REQUEST_ROOT, node_request.id)
if self.client.exists(path):
return True
return False
def storeNodeRequest(self, node_request):
'''Store the node request.
The request is expected to already exist and is updated in its
entirety.
:param NodeRequest node_request: The request to update.
'''
path = '%s/%s' % (self.REQUEST_ROOT, node_request.id)
self.client.set(path, self._dictToStr(node_request.toDict()))
def updateNodeRequest(self, node_request, data=None):
'''Refresh an existing node request.
:param NodeRequest node_request: The request to update.
:param dict data: The data to use; query ZK if absent.
'''
if data is None:
path = '%s/%s' % (self.REQUEST_ROOT, node_request.id)
data, stat = self.client.get(path)
data = self._strToDict(data)
request_nodes = list(node_request.nodeset.getNodes())
for i, nodeid in enumerate(data.get('nodes', [])):
request_nodes[i].id = nodeid
self.updateNode(request_nodes[i])
node_request.updateFromDict(data)
def storeNode(self, node):
'''Store the node.
The node is expected to already exist and is updated in its
entirety.
:param Node node: The node to update.
'''
path = '%s/%s' % (self.NODE_ROOT, node.id)
self.client.set(path, self._dictToStr(node.toDict()))
def updateNode(self, node):
'''Refresh an existing node.
:param Node node: The node to update.
'''
node_path = '%s/%s' % (self.NODE_ROOT, node.id)
node_data, node_stat = self.client.get(node_path)
node_data = self._strToDict(node_data)
node.updateFromDict(node_data)
def lockNode(self, node, blocking=True, timeout=None):
'''
Lock a node.
This should be called as soon as a request is fulfilled and
the lock held for as long as the node is in-use. It can be
used by nodepool to detect if Zuul has gone offline and the
node should be reclaimed.
:param Node node: The node which should be locked.
'''
lock_path = '%s/%s/lock' % (self.NODE_ROOT, node.id)
try:
lock = Lock(self.client, lock_path)
have_lock = lock.acquire(blocking, timeout)
except kze.LockTimeout:
raise LockException("Timeout trying to acquire lock %s" %
lock_path)
# If we aren't blocking, it's possible we didn't get the lock
# because someone else has it.
if not have_lock:
raise LockException("Did not get lock on %s" % lock_path)
node.lock = lock
def unlockNode(self, node):
'''
Unlock a node.
The node must already have been locked.
:param Node node: The node which should be unlocked.
'''
if node.lock is None:
raise LockException("Node %s does not hold a lock" % (node, ))
node.lock.release()
node.lock = None
def lockNodeRequest(self, request, blocking=True, timeout=None):
'''
Lock a node request.
This will set the `lock` attribute of the request object when the
lock is successfully acquired.
:param NodeRequest request: The request to lock.
:param bool blocking: Whether or not to block on trying to
acquire the lock
:param int timeout: When blocking, how long to wait for the lock
to get acquired. None, the default, waits forever.
:raises: TimeoutException if we failed to acquire the lock when
blocking with a timeout. ZKLockException if we are not blocking
and could not get the lock, or a lock is already held.
'''
path = "%s/%s" % (self.REQUEST_LOCK_ROOT, request.id)
try:
lock = Lock(self.client, path)
have_lock = lock.acquire(blocking, timeout)
except kze.LockTimeout:
raise LockException("Timeout trying to acquire lock %s" % path)
except kze.NoNodeError:
have_lock = False
self.log.error("Request not found for locking: %s", request)
# If we aren't blocking, it's possible we didn't get the lock
# because someone else has it.
if not have_lock:
raise LockException("Did not get lock on %s" % path)
request.lock = lock
self.updateNodeRequest(request)
def unlockNodeRequest(self, request):
'''
Unlock a node request.
The request must already have been locked.
:param NodeRequest request: The request to unlock.
:raises: ZKLockException if the request is not currently locked.
'''
if request.lock is None:
raise LockException("Request %s does not hold a lock" % request)
request.lock.release()
request.lock = None
def heldNodeCount(self, autohold_key):
'''
Count the number of nodes being held for the given tenant/project/job.
:param set autohold_key: A set with the tenant/project/job names.
'''
identifier = " ".join(autohold_key)
try:
nodes = self.client.get_children(self.NODE_ROOT)
except kze.NoNodeError:
return 0
count = 0
for nodeid in nodes:
node_path = '%s/%s' % (self.NODE_ROOT, nodeid)
try:
node_data, node_stat = self.client.get(node_path)
except kze.NoNodeError:
# Node got removed on us. Just ignore.
continue
if not node_data:
self.log.warning("Node ID %s has no data", nodeid)
continue
node_data = self._strToDict(node_data)
if (node_data['state'] == zuul.model.STATE_HOLD
and node_data.get('hold_job') == identifier):
count += 1
return count
# Copy of nodepool/zk.py begins here
NODE_ROOT = "/nodepool/nodes"
LAUNCHER_ROOT = "/nodepool/launchers"
def _bytesToDict(self, data):
return json.loads(data.decode('utf8'))
def _launcherPath(self, launcher):
return "%s/%s" % (self.LAUNCHER_ROOT, launcher)
def _nodePath(self, node):
return "%s/%s" % (self.NODE_ROOT, node)
def getRegisteredLaunchers(self):
'''
Get a list of all launchers that have registered with ZooKeeper.
:returns: A list of Launcher objects, or empty list if none are found.
'''
try:
launcher_ids = self.client.get_children(self.LAUNCHER_ROOT)
except kze.NoNodeError:
return []
objs = []
for launcher in launcher_ids:
path = self._launcherPath(launcher)
try:
data, _ = self.client.get(path)
except kze.NoNodeError:
# launcher disappeared
continue
objs.append(Launcher.fromDict(self._bytesToDict(data)))
return objs
def getNodes(self):
'''
Get the current list of all nodes.
:returns: A list of nodes.
'''
try:
return self.client.get_children(self.NODE_ROOT)
except kze.NoNodeError:
return []
def getNode(self, node):
'''
Get the data for a specific node.
:param str node: The node ID.
:returns: The node data, or None if the node was not found.
'''
path = self._nodePath(node)
try:
data, stat = self.client.get(path)
except kze.NoNodeError:
return None
if not data:
return None
d = self._bytesToDict(data)
d['id'] = node
return d
def nodeIterator(self):
'''
Utility generator method for iterating through all nodes.
'''
for node_id in self.getNodes():
node = self.getNode(node_id)
if node:
yield node
def getHoldRequests(self):
'''
Get the current list of all hold requests.
'''
try:
return sorted(self.client.get_children(self.HOLD_REQUEST_ROOT))
except kze.NoNodeError:
return []
def getHoldRequest(self, hold_request_id):
path = self.HOLD_REQUEST_ROOT + "/" + hold_request_id
try:
data, stat = self.client.get(path)
except kze.NoNodeError:
return None
if not data:
return None
obj = zuul.model.HoldRequest.fromDict(self._strToDict(data))
obj.id = hold_request_id
obj.stat = stat
return obj
def storeHoldRequest(self, hold_request):
'''
Create or update a hold request.
If this is a new request with no value for the `id` attribute of the
passed in request, then `id` will be set with the unique request
identifier after successful creation.
:param HoldRequest hold_request: Object representing the hold request.
'''
if hold_request.id is None:
path = self.client.create(self.HOLD_REQUEST_ROOT + "/",
value=hold_request.serialize(),
sequence=True,
makepath=True)
hold_request.id = path.split('/')[-1]
else:
path = self.HOLD_REQUEST_ROOT + "/" + hold_request.id
self.client.set(path, hold_request.serialize())
def _markHeldNodesAsUsed(self, hold_request):
'''
Changes the state for each held node for the hold request to 'used'.
:returns: True if all nodes marked USED, False otherwise.
'''
def getHeldNodeIDs(request):
node_ids = []
for data in request.nodes:
# TODO(Shrews): Remove type check at some point.
# When autoholds were initially changed to be stored in ZK,
# the node IDs were originally stored as a list of strings.
# A later change embedded them within a dict. Handle both
# cases here to deal with the upgrade.
if isinstance(data, dict):
node_ids += data['nodes']
else:
node_ids.append(data)
return node_ids
failure = False
for node_id in getHeldNodeIDs(hold_request):
node = self.getNode(node_id)
if not node or node['state'] == zuul.model.STATE_USED:
continue
node['state'] = zuul.model.STATE_USED
name = None
label = None
if 'name' in node:
name = node['name']
if 'label' in node:
label = node['label']
node_obj = zuul.model.Node(name, label)
node_obj.updateFromDict(node)
try:
self.lockNode(node_obj, blocking=False)
self.storeNode(node_obj)
except Exception:
self.log.exception(
"Cannot change HELD node state to USED "
"for node %s in request %s", node_obj.id, hold_request.id)
failure = True
finally:
try:
if node_obj.lock:
self.unlockNode(node_obj)
except Exception:
self.log.exception(
"Failed to unlock HELD node %s for request %s",
node_obj.id, hold_request.id)
return not failure
def deleteHoldRequest(self, hold_request):
'''
Delete a hold request.
:param HoldRequest hold_request: Object representing the hold request.
'''
if not self._markHeldNodesAsUsed(hold_request):
self.log.info(
"Unable to delete hold request %s because "
"not all nodes marked as USED.", hold_request.id)
return
path = self.HOLD_REQUEST_ROOT + "/" + hold_request.id
try:
self.client.delete(path, recursive=True)
except kze.NoNodeError:
pass
def lockHoldRequest(self, request, blocking=True, timeout=None):
'''
Lock a node request.
This will set the `lock` attribute of the request object when the
lock is successfully acquired.
:param HoldRequest request: The hold request to lock.
'''
if not request.id:
raise LockException(
"Hold request without an ID cannot be locked: %s" % request)
path = "%s/%s/lock" % (self.HOLD_REQUEST_ROOT, request.id)
try:
lock = Lock(self.client, path)
have_lock = lock.acquire(blocking, timeout)
except kze.LockTimeout:
raise LockException("Timeout trying to acquire lock %s" % path)
# If we aren't blocking, it's possible we didn't get the lock
# because someone else has it.
if not have_lock:
raise LockException("Did not get lock on %s" % path)
request.lock = lock
def unlockHoldRequest(self, request):
'''
Unlock a hold request.
The request must already have been locked.
:param HoldRequest request: The request to unlock.
:raises: ZKLockException if the request is not currently locked.
'''
if request.lock is None:
raise LockException("Request %s does not hold a lock" % request)
request.lock.release()
request.lock = None
class ZooKeeper(AbstractDCS):
def __init__(self, name, config):
super(ZooKeeper, self).__init__(name, config)
hosts = config.get('hosts', [])
if isinstance(hosts, list):
hosts = ','.join(hosts)
self.exhibitor = None
if 'exhibitor' in config:
exhibitor = config['exhibitor']
interval = exhibitor.get('poll_interval', 300)
self.exhibitor = ExhibitorEnsembleProvider(exhibitor['hosts'],
exhibitor['port'],
poll_interval=interval)
hosts = self.exhibitor.zookeeper_hosts
self.client = KazooClient(hosts=hosts,
timeout=(config.get('session_timeout', None)
or 30),
command_retry={
'deadline':
(config.get('reconnect_timeout', None)
or 10),
'max_delay':
1,
'max_tries':
-1
},
connection_retry={
'max_delay': 1,
'max_tries': -1
})
self.client.add_listener(self.session_listener)
self.cluster_event = self.client.handler.event_object()
self.fetch_cluster = True
self.members = []
self.leader = None
self.last_leader_operation = 0
self.client.start(None)
def session_listener(self, state):
if state in [KazooState.SUSPENDED, KazooState.LOST]:
self.cluster_watcher(None)
def cluster_watcher(self, event):
self.fetch_cluster = True
self.cluster_event.set()
def get_node(self, name, watch=None):
try:
return self.client.get(self.client_path(name), watch)
except NoNodeError:
pass
except:
logger.exception('get_node')
return None
@staticmethod
def member(name, value, znode):
conn_url, api_url = parse_connection_string(value)
return Member(znode.mzxid, name, conn_url, api_url, None, None)
def load_members(self):
members = []
for member in self.client.get_children(self.client_path('/members'),
self.cluster_watcher):
data = self.get_node('/members/' + member)
if data is not None:
members.append(self.member(member, *data))
return members
def _inner_load_cluster(self):
self.cluster_event.clear()
leader = self.get_node('/leader', self.cluster_watcher)
self.members = self.load_members()
if leader:
if leader[0] == self._name:
client_id = self.client.client_id
if client_id is not None and client_id[0] != leader[
1].ephemeralOwner:
logger.info(
'I am leader but not owner of the session. Removing leader node'
)
self.client.delete(self.client_path('/leader'))
leader = None
if leader:
for member in self.members:
if member.name == leader[0]:
leader = member
self.fetch_cluster = False
break
if not isinstance(leader, Member):
leader = Member(-1, leader, None, None, None, None)
self.leader = leader
if self.fetch_cluster:
last_leader_operation = self.get_node('/optime/leader')
if last_leader_operation:
self.last_leader_operation = int(last_leader_operation[0])
def get_cluster(self):
if self.exhibitor and self.exhibitor.poll():
self.client.set_hosts(self.exhibitor.zookeeper_hosts)
if self.fetch_cluster:
try:
self.client.retry(self._inner_load_cluster)
except:
logger.exception('get_cluster')
self.session_listener(KazooState.LOST)
raise ZooKeeperError('ZooKeeper in not responding properly')
return Cluster(True, self.leader, self.last_leader_operation,
self.members)
def _create(self, path, value, **kwargs):
try:
self.client.retry(self.client.create, self.client_path(path),
value, **kwargs)
return True
except:
return False
def attempt_to_acquire_leader(self):
ret = self._create('/leader',
self._name,
makepath=True,
ephemeral=True)
ret or logger.info('Could not take out TTL lock')
return ret
def race(self, path):
return self._create(path, self._name, makepath=True)
def touch_member(self, connection_string, ttl=None):
for m in self.members:
if m.name == self._name:
return True
path = self.client_path('/members/' + self._name)
try:
self.client.retry(self.client.create,
path,
connection_string,
makepath=True,
ephemeral=True)
return True
except NodeExistsError:
try:
self.client.retry(self.client.delete, path)
self.client.retry(self.client.create,
path,
connection_string,
makepath=True,
ephemeral=True)
return True
except:
logger.exception('touch_member')
return False
def take_leader(self):
return self.attempt_to_acquire_leader()
def update_leader(self, state_handler):
last_operation = state_handler.last_operation()
if last_operation != self.last_leader_operation:
self.last_leader_operation = last_operation
path = self.client_path('/optime/leader')
try:
self.client.retry(self.client.set, path, last_operation)
except NoNodeError:
try:
self.client.retry(self.client.create,
path,
last_operation,
makepath=True)
except:
logger.exception('Failed to create %s', path)
except:
logger.exception('Failed to update %s', path)
return True
def delete_leader(self):
if isinstance(self.leader, Member) and self.leader.name == self._name:
self.client.delete(self.client_path('/leader'))
def sleep(self, timeout):
self.cluster_event.wait(timeout)
if self.cluster_event.isSet():
self.fetch_cluster = True
class ZooKeeper(AbstractDCS):
def __init__(self, name, config):
super(ZooKeeper, self).__init__(name, config)
hosts = config.get('hosts', [])
if isinstance(hosts, list):
hosts = ','.join(hosts)
self.exhibitor = None
if 'exhibitor' in config:
exhibitor = config['exhibitor']
interval = exhibitor.get('poll_interval', 300)
self.exhibitor = ExhibitorEnsembleProvider(exhibitor['hosts'], exhibitor['port'], poll_interval=interval)
hosts = self.exhibitor.zookeeper_hosts
self.client = KazooClient(hosts=hosts,
timeout=(config.get('session_timeout', None) or 30),
command_retry={
'deadline': (config.get('reconnect_timeout', None) or 10),
'max_delay': 1,
'max_tries': -1},
connection_retry={'max_delay': 1, 'max_tries': -1})
self.client.add_listener(self.session_listener)
self.cluster_event = self.client.handler.event_object()
self.fetch_cluster = True
self.members = []
self.leader = None
self.last_leader_operation = 0
self.client.start(None)
def session_listener(self, state):
if state in [KazooState.SUSPENDED, KazooState.LOST]:
self.cluster_watcher(None)
def cluster_watcher(self, event):
self.fetch_cluster = True
self.cluster_event.set()
def get_node(self, name, watch=None):
try:
return self.client.get(self.client_path(name), watch)
except NoNodeError:
pass
except:
logger.exception('get_node')
return None
@staticmethod
def member(name, value, znode):
conn_url, api_url = parse_connection_string(value)
return Member(znode.mzxid, name, conn_url, api_url, None, None)
def load_members(self):
members = []
for member in self.client.get_children(self.client_path('/members'), self.cluster_watcher):
data = self.get_node('/members/' + member)
if data is not None:
members.append(self.member(member, *data))
return members
def _inner_load_cluster(self):
self.cluster_event.clear()
leader = self.get_node('/leader', self.cluster_watcher)
self.members = self.load_members()
if leader:
if leader[0] == self._name:
client_id = self.client.client_id
if client_id is not None and client_id[0] != leader[1].ephemeralOwner:
logger.info('I am leader but not owner of the session. Removing leader node')
self.client.delete(self.client_path('/leader'))
leader = None
if leader:
for member in self.members:
if member.name == leader[0]:
leader = member
self.fetch_cluster = False
break
if not isinstance(leader, Member):
leader = Member(-1, leader, None, None, None, None)
self.leader = leader
if self.fetch_cluster:
last_leader_operation = self.get_node('/optime/leader')
if last_leader_operation:
self.last_leader_operation = int(last_leader_operation[0])
def get_cluster(self):
if self.exhibitor and self.exhibitor.poll():
self.client.set_hosts(self.exhibitor.zookeeper_hosts)
if self.fetch_cluster:
try:
self.client.retry(self._inner_load_cluster)
except:
logger.exception('get_cluster')
self.session_listener(KazooState.LOST)
raise ZooKeeperError('ZooKeeper in not responding properly')
return Cluster(True, self.leader, self.last_leader_operation, self.members)
def _create(self, path, value, **kwargs):
try:
self.client.retry(self.client.create, self.client_path(path), value, **kwargs)
return True
except:
return False
def attempt_to_acquire_leader(self):
ret = self._create('/leader', self._name, makepath=True, ephemeral=True)
ret or logger.info('Could not take out TTL lock')
return ret
def race(self, path):
return self._create(path, self._name, makepath=True)
def touch_member(self, connection_string, ttl=None):
for m in self.members:
if m.name == self._name:
return True
path = self.client_path('/members/' + self._name)
try:
self.client.retry(self.client.create, path, connection_string, makepath=True, ephemeral=True)
return True
except NodeExistsError:
try:
self.client.retry(self.client.delete, path)
self.client.retry(self.client.create, path, connection_string, makepath=True, ephemeral=True)
return True
except:
logger.exception('touch_member')
return False
def take_leader(self):
return self.attempt_to_acquire_leader()
def update_leader(self, state_handler):
last_operation = state_handler.last_operation()
if last_operation != self.last_leader_operation:
self.last_leader_operation = last_operation
path = self.client_path('/optime/leader')
try:
self.client.retry(self.client.set, path, last_operation)
except NoNodeError:
try:
self.client.retry(self.client.create, path, last_operation, makepath=True)
except:
logger.exception('Failed to create %s', path)
except:
logger.exception('Failed to update %s', path)
return True
def delete_leader(self):
if isinstance(self.leader, Member) and self.leader.name == self._name:
self.client.delete(self.client_path('/leader'))
def sleep(self, timeout):
self.cluster_event.wait(timeout)
if self.cluster_event.isSet():
self.fetch_cluster = True
class _ZookeeperProxy(object):
def __init__(self, address_provider: AddressListProvider, prefix: str):
self.address_provider = address_provider
self.async_counter = WaitingCounter(limit=100)
self.conn_str = None
self.client = None
self.prefix = prefix
self.hosts_cache = SlowlyUpdatedCache(
self.address_provider.get_latest_address,
self._update_hosts,
30, # Refresh every 30 seconds
3 * 60) # Update only after 180 seconds of stability
def _update_hosts(self, value):
hosts, port = value
if hosts:
self.conn_str = ','.join(['{}:{}'.format(h, port) for h in hosts]) + self.prefix
if self.client is None:
self.client = KazooClient(hosts=self.conn_str,
command_retry={'deadline': 120, 'max_delay': 1, 'max_tries': -1},
connection_retry={'max_delay': 1, 'max_tries': -1})
self.client.add_listener(self.session_listener)
else:
self.client.stop()
self.client.set_hosts(self.conn_str)
self.client.start()
def terminate(self):
if self.client:
self.client.stop()
def session_listener(self, state):
pass
def get_conn_str(self):
return self.conn_str
def get(self, *params):
self.hosts_cache.touch()
return self.client.retry(self.client.get, *params)
def get_async(self, *params):
# Exhibitor is not polled here and it's totally fine!
self.async_counter.increment()
try:
i_async = self.client.get_async(*params)
i_async.rawlink(self._decrement)
return i_async
except Exception as e:
self._decrement()
raise e
def _decrement(self, *args, **kwargs):
self.async_counter.decrement()
def set(self, *args, **kwargs):
self.hosts_cache.touch()
return self.client.retry(self.client.set, *args, **kwargs)
def create(self, *args, **kwargs):
self.hosts_cache.touch()
return self.client.retry(self.client.create, *args, **kwargs)
def delete(self, *args, **kwargs):
self.hosts_cache.touch()
try:
return self.client.retry(self.client.delete, *args, **kwargs)
except NoNodeError:
pass
def get_children(self, *args, **kwargs):
self.hosts_cache.touch()
try:
return self.client.retry(self.client.get_children, *args, **kwargs)
except NoNodeError:
return []
def take_lock(self, *args, **kwargs):
while True:
try:
self.hosts_cache.touch()
return self.client.Lock(*args, **kwargs)
except Exception as e:
_LOG.error('Failed to obtain lock for exhibitor, retrying', exc_info=e)
class ZooKeeper(object):
'''
Class implementing the ZooKeeper interface.
This class uses the facade design pattern to keep common interaction
with the ZooKeeper API simple and consistent for the caller, and
limits coupling between objects. It allows for more complex interactions
by providing direct access to the client connection when needed (though
that is discouraged). It also provides for a convenient entry point for
testing only ZooKeeper interactions.
'''
log = logging.getLogger("zuul.zk.ZooKeeper")
REQUEST_ROOT = '/nodepool/requests'
REQUEST_LOCK_ROOT = "/nodepool/requests-lock"
NODE_ROOT = '/nodepool/nodes'
# Log zookeeper retry every 10 seconds
retry_log_rate = 10
def __init__(self):
'''
Initialize the ZooKeeper object.
'''
self.client = None
self._became_lost = False
self._last_retry_log = 0
def _dictToStr(self, data):
return json.dumps(data).encode('utf8')
def _strToDict(self, data):
return json.loads(data.decode('utf8'))
def _connection_listener(self, state):
'''
Listener method for Kazoo connection state changes.
.. warning:: This method must not block.
'''
if state == KazooState.LOST:
self.log.debug("ZooKeeper connection: LOST")
self._became_lost = True
elif state == KazooState.SUSPENDED:
self.log.debug("ZooKeeper connection: SUSPENDED")
else:
self.log.debug("ZooKeeper connection: CONNECTED")
@property
def connected(self):
return self.client.state == KazooState.CONNECTED
@property
def suspended(self):
return self.client.state == KazooState.SUSPENDED
@property
def lost(self):
return self.client.state == KazooState.LOST
@property
def didLoseConnection(self):
return self._became_lost
def resetLostFlag(self):
self._became_lost = False
def logConnectionRetryEvent(self):
now = time.monotonic()
if now - self._last_retry_log >= self.retry_log_rate:
self.log.warning("Retrying zookeeper connection")
self._last_retry_log = now
def connect(self, hosts, read_only=False, timeout=10.0):
'''
Establish a connection with ZooKeeper cluster.
Convenience method if a pre-existing ZooKeeper connection is not
supplied to the ZooKeeper object at instantiation time.
:param str hosts: Comma-separated list of hosts to connect to (e.g.
127.0.0.1:2181,127.0.0.1:2182,[::1]:2183).
:param bool read_only: If True, establishes a read-only connection.
:param float timeout: The ZooKeeper session timeout, in
seconds (default: 10.0).
'''
if self.client is None:
self.client = KazooClient(hosts=hosts, read_only=read_only,
timeout=timeout)
self.client.add_listener(self._connection_listener)
# Manually retry initial connection attempt
while True:
try:
self.client.start(1)
break
except KazooTimeoutError:
self.logConnectionRetryEvent()
def disconnect(self):
'''
Close the ZooKeeper cluster connection.
You should call this method if you used connect() to establish a
cluster connection.
'''
if self.client is not None and self.client.connected:
self.client.stop()
self.client.close()
self.client = None
def resetHosts(self, hosts):
'''
Reset the ZooKeeper cluster connection host list.
:param str hosts: Comma-separated list of hosts to connect to (e.g.
127.0.0.1:2181,127.0.0.1:2182,[::1]:2183).
'''
if self.client is not None:
self.client.set_hosts(hosts=hosts)
def submitNodeRequest(self, node_request, watcher):
'''
Submit a request for nodes to Nodepool.
:param NodeRequest node_request: A NodeRequest with the
contents of the request.
:param callable watcher: A callable object that will be
invoked each time the request is updated. It is called
with two arguments: (node_request, deleted) where
node_request is the same argument passed to this method,
and deleted is a boolean which is True if the node no
longer exists (notably, this will happen on disconnection
from ZooKeeper). The watcher should return False when
further updates are no longer necessary.
'''
node_request.created_time = time.time()
data = node_request.toDict()
path = '{}/{:0>3}-'.format(self.REQUEST_ROOT, node_request.priority)
path = self.client.create(path, self._dictToStr(data),
makepath=True,
sequence=True, ephemeral=True)
reqid = path.split("/")[-1]
node_request.id = reqid
def callback(data, stat):
if data:
self.updateNodeRequest(node_request, data)
deleted = (data is None) # data *are* none
return watcher(node_request, deleted)
self.client.DataWatch(path, callback)
def deleteNodeRequest(self, node_request):
'''
Delete a request for nodes.
:param NodeRequest node_request: A NodeRequest with the
contents of the request.
'''
path = '%s/%s' % (self.REQUEST_ROOT, node_request.id)
try:
self.client.delete(path)
except kze.NoNodeError:
pass
def nodeRequestExists(self, node_request):
'''
See if a NodeRequest exists in ZooKeeper.
:param NodeRequest node_request: A NodeRequest to verify.
:returns: True if the request exists, False otherwise.
'''
path = '%s/%s' % (self.REQUEST_ROOT, node_request.id)
if self.client.exists(path):
return True
return False
def storeNodeRequest(self, node_request):
'''Store the node request.
The request is expected to already exist and is updated in its
entirety.
:param NodeRequest node_request: The request to update.
'''
path = '%s/%s' % (self.REQUEST_ROOT, node_request.id)
self.client.set(path, self._dictToStr(node_request.toDict()))
def updateNodeRequest(self, node_request, data=None):
'''Refresh an existing node request.
:param NodeRequest node_request: The request to update.
:param dict data: The data to use; query ZK if absent.
'''
if data is None:
path = '%s/%s' % (self.REQUEST_ROOT, node_request.id)
data, stat = self.client.get(path)
data = self._strToDict(data)
request_nodes = list(node_request.nodeset.getNodes())
for i, nodeid in enumerate(data.get('nodes', [])):
request_nodes[i].id = nodeid
self.updateNode(request_nodes[i])
node_request.updateFromDict(data)
def storeNode(self, node):
'''Store the node.
The node is expected to already exist and is updated in its
entirety.
:param Node node: The node to update.
'''
path = '%s/%s' % (self.NODE_ROOT, node.id)
self.client.set(path, self._dictToStr(node.toDict()))
def updateNode(self, node):
'''Refresh an existing node.
:param Node node: The node to update.
'''
node_path = '%s/%s' % (self.NODE_ROOT, node.id)
node_data, node_stat = self.client.get(node_path)
node_data = self._strToDict(node_data)
node.updateFromDict(node_data)
def lockNode(self, node, blocking=True, timeout=None):
'''
Lock a node.
This should be called as soon as a request is fulfilled and
the lock held for as long as the node is in-use. It can be
used by nodepool to detect if Zuul has gone offline and the
node should be reclaimed.
:param Node node: The node which should be locked.
'''
lock_path = '%s/%s/lock' % (self.NODE_ROOT, node.id)
try:
lock = Lock(self.client, lock_path)
have_lock = lock.acquire(blocking, timeout)
except kze.LockTimeout:
raise LockException(
"Timeout trying to acquire lock %s" % lock_path)
# If we aren't blocking, it's possible we didn't get the lock
# because someone else has it.
if not have_lock:
raise LockException("Did not get lock on %s" % lock_path)
node.lock = lock
def unlockNode(self, node):
'''
Unlock a node.
The node must already have been locked.
:param Node node: The node which should be unlocked.
'''
if node.lock is None:
raise LockException("Node %s does not hold a lock" % (node,))
node.lock.release()
node.lock = None
def lockNodeRequest(self, request, blocking=True, timeout=None):
'''
Lock a node request.
This will set the `lock` attribute of the request object when the
lock is successfully acquired.
:param NodeRequest request: The request to lock.
:param bool blocking: Whether or not to block on trying to
acquire the lock
:param int timeout: When blocking, how long to wait for the lock
to get acquired. None, the default, waits forever.
:raises: TimeoutException if we failed to acquire the lock when
blocking with a timeout. ZKLockException if we are not blocking
and could not get the lock, or a lock is already held.
'''
path = "%s/%s" % (self.REQUEST_LOCK_ROOT, request.id)
try:
lock = Lock(self.client, path)
have_lock = lock.acquire(blocking, timeout)
except kze.LockTimeout:
raise LockException(
"Timeout trying to acquire lock %s" % path)
except kze.NoNodeError:
have_lock = False
self.log.error("Request not found for locking: %s", request)
# If we aren't blocking, it's possible we didn't get the lock
# because someone else has it.
if not have_lock:
raise LockException("Did not get lock on %s" % path)
request.lock = lock
self.updateNodeRequest(request)
def unlockNodeRequest(self, request):
'''
Unlock a node request.
The request must already have been locked.
:param NodeRequest request: The request to unlock.
:raises: ZKLockException if the request is not currently locked.
'''
if request.lock is None:
raise LockException(
"Request %s does not hold a lock" % request)
request.lock.release()
request.lock = None
def heldNodeCount(self, autohold_key):
'''
Count the number of nodes being held for the given tenant/project/job.
:param set autohold_key: A set with the tenant/project/job names.
'''
identifier = " ".join(autohold_key)
try:
nodes = self.client.get_children(self.NODE_ROOT)
except kze.NoNodeError:
return 0
count = 0
for nodeid in nodes:
node_path = '%s/%s' % (self.NODE_ROOT, nodeid)
try:
node_data, node_stat = self.client.get(node_path)
except kze.NoNodeError:
# Node got removed on us. Just ignore.
continue
if not node_data:
self.log.warning("Node ID %s has no data", nodeid)
continue
node_data = self._strToDict(node_data)
if (node_data['state'] == zuul.model.STATE_HOLD and
node_data.get('hold_job') == identifier):
count += 1
return count
# Copy of nodepool/zk.py begins here
NODE_ROOT = "/nodepool/nodes"
LAUNCHER_ROOT = "/nodepool/launchers"
def _bytesToDict(self, data):
return json.loads(data.decode('utf8'))
def _launcherPath(self, launcher):
return "%s/%s" % (self.LAUNCHER_ROOT, launcher)
def _nodePath(self, node):
return "%s/%s" % (self.NODE_ROOT, node)
def getRegisteredLaunchers(self):
'''
Get a list of all launchers that have registered with ZooKeeper.
:returns: A list of Launcher objects, or empty list if none are found.
'''
try:
launcher_ids = self.client.get_children(self.LAUNCHER_ROOT)
except kze.NoNodeError:
return []
objs = []
for launcher in launcher_ids:
path = self._launcherPath(launcher)
try:
data, _ = self.client.get(path)
except kze.NoNodeError:
# launcher disappeared
continue
objs.append(Launcher.fromDict(self._bytesToDict(data)))
return objs
def getNodes(self):
'''
Get the current list of all nodes.
:returns: A list of nodes.
'''
try:
return self.client.get_children(self.NODE_ROOT)
except kze.NoNodeError:
return []
def getNode(self, node):
'''
Get the data for a specific node.
:param str node: The node ID.
:returns: The node data, or None if the node was not found.
'''
path = self._nodePath(node)
try:
data, stat = self.client.get(path)
except kze.NoNodeError:
return None
if not data:
return None
d = self._bytesToDict(data)
d['id'] = node
return d
def nodeIterator(self):
'''
Utility generator method for iterating through all nodes.
'''
for node_id in self.getNodes():
node = self.getNode(node_id)
if node:
yield node
class MainWindow(QMainWindow, ui_MainWindow.Ui_MainWindow):
mainWriteGui = pyqtSignal(str)
@catchExept
def __init__(self):
super().__init__()
self.setupUi(self)
self.zk = KazooClient()
self.zkTimer = QTimer(self)
self.zkTimer.setInterval(100)
self.zkTimer.timeout.connect(self.zkTimeout)
self.zkStartThread = threading.Thread(target=self.zkConnect)
self.msgBox = QMessageBox(QMessageBox.NoIcon, "Connection",
"Connecting...", QMessageBox.Cancel, self)
self.treeWidget.itemClicked.connect(self.itemClicked)
self.treeWidget.itemDoubleClicked.connect(self.itemOpen)
self.tabWidget.tabCloseRequested.connect(self.closeTab)
self.actionConnect.triggered.connect(self.msgBox.show)
self.actionConnect.triggered.connect(self.zkStartThread.start)
self.actionConnect.triggered.connect(self.zkTimer.start)
self.actionDisconnect.triggered.connect(self.zkDisconnect)
self.actionACLVersion.triggered.connect(self.aclVersion)
self.actionCreated.triggered.connect(self.created)
self.actionChildrenCount.triggered.connect(self.childrenCount)
self.actionDataLength.triggered.connect(self.dataLength)
self.actionLastModified.triggered.connect(self.lastModified)
self.actionLastModifiedTransactionId.triggered.connect(
self.lastModifiedTransactionId)
self.actionOwnerSessionId.triggered.connect(self.ownerSessionId)
self.actionVersion.triggered.connect(self.version)
self.actionCreationTransactionId.triggered.connect(
self.creationTransactionId)
self.actionChangeServerAddress.triggered.connect(
self.changeServerAddress)
self.msgBox.rejected.connect(self.zkTimer.stop)
self.msgBox.rejected.connect(self.msgBox.hide)
self.msgBox.rejected.connect(self.zkDisconnect)
self.mainWriteGui.connect(self.slotMainWriteGui)
self.log.setCenterOnScroll(True)
self.dialog = SelectorDialog(self)
class PlainTextWidgetHandler:
def __init__(self, logToWriteGui):
self.logToWriteGui = logToWriteGui
def write(self, text):
self.logToWriteGui(text)
def flush(self):
pass
logging.basicConfig(format='%(asctime)s.%(msecs)d: %(message)s',
datefmt='%H:%M:%S',
level=logging.DEBUG,
handlers=[
logging.StreamHandler(
PlainTextWidgetHandler(
self.logToWriteGui)),
logging.StreamHandler(sys.stderr)
])
self.treeWidget.setColumnCount(1)
self.treeWidget.sortByColumn(0, Qt.AscendingOrder)
l = self.msgBox.layout()
progress = QProgressBar()
progress.setMaximum(0)
progress.setMinimum(0)
l.addWidget(progress, l.rowCount() - 2, 1, 1, l.columnCount())
self.actionConnect.setEnabled(False)
if os.path.exists("config.txt"):
with open("config.txt", "r") as f:
prelines = [string.strip() for string in f.readlines()]
lines = [prelines[0]]
for i in range(1, len(prelines)):
if prelines[i -
1] != prelines[i] and prelines[i] not in lines:
lines.append(prelines[i])
lines = list(filter(None, lines))
self.dialog.comboBox.addItems(lines)
self.actionConnect.setEnabled(True)
@pyqtSlot(str)
def slotMainWriteGui(self, text):
self.log.ensureCursorVisible()
self.log.textCursor().insertText(text)
def logToWriteGui(self, text):
self.mainWriteGui.emit(text)
@catchExept
def getCurrentStat(self):
_, stat = self.zk.get(self.treeWidget.currentItem().text(1))
return stat
@catchExept
@pyqtSlot()
def aclVersion(self):
self.print("ACL version: %s" % self.getCurrentStat().acl_version)
@catchExept
@pyqtSlot()
def created(self):
self.print("Created: %s" % self.getCurrentStat().created)
@catchExept
@pyqtSlot()
def childrenCount(self):
self.print("Children count: %s" % self.getCurrentStat().children_count)
@catchExept
@pyqtSlot()
def dataLength(self):
self.print("Data length: %s" % self.getCurrentStat().data_length)
@catchExept
@pyqtSlot()
def lastModified(self):
self.print("Last modified: %s" % self.getCurrentStat().last_modified)
@catchExept
@pyqtSlot()
def lastModifiedTransactionId(self):
self.print("Last modified transactionId: %s" %
self.getCurrentStat().last_modified_transaction_id)
@catchExept
@pyqtSlot()
def ownerSessionId(self):
self.print("Owner sessionId: %s" %
self.getCurrentStat().owner_session_id)
@catchExept
@pyqtSlot()
def version(self):
self.print("Version: %s" % self.getCurrentStat().version)
@catchExept
@pyqtSlot()
def creationTransactionId(self):
self.print("Creation transactionId: %s" %
self.getCurrentStat().creation_transaction_id)
@catchExept
@pyqtSlot()
def changeServerAddress(self):
code = self.dialog.exec_()
if code == QDialog.Accepted:
text = self.currentHost()
with open("config.txt", "w") as f:
hosts = [
self.dialog.comboBox.itemText(s)
for s in range(self.dialog.comboBox.count())
if text != self.dialog.comboBox.itemText(s)
]
self.dialog.comboBox.clear()
self.dialog.comboBox.addItems(hosts)
hosts.insert(0, text)
f.write('\n'.join(hosts))
if text != "":
if text != self.dialog.comboBox.itemText(0):
self.dialog.comboBox.insertItem(0, text)
self.dialog.comboBox.setCurrentText(text)
self.print("Current host changed to %s" % self.currentHost())
self.actionConnect.setEnabled(True)
@catchExept
@pyqtSlot()
def zkDisconnect(self):
self.tabWidget.clear()
self.treeWidget.clear()
self.zk.stop()
self.zk.close()
self.actionDisconnect.setEnabled(False)
self.menuFileInfo.setEnabled(False)
self.actionConnect.setEnabled(True)
self.actionChangeServerAddress.setEnabled(True)
self.zkStartThread = threading.Thread(target=self.zkConnect)
self.actionConnect.triggered.connect(self.zkStartThread.start)
@catchExept
@pyqtSlot()
def zkTimeout(self):
if self.zk.connected:
self.zkConnected()
self.zkTimer.stop()
@catchExept
def zkConnected(self):
self.msgBox.hide()
self.init()
self.menuFileInfo.setEnabled(True)
self.actionDisconnect.setEnabled(True)
self.actionConnect.setEnabled(False)
self.actionChangeServerAddress.setEnabled(False)
@catchExept
def zkConnect(self):
self.zk.set_hosts(self.currentHost())
self.zk.add_listener(self.my_listener)
try:
self.zk.start_async()
except Exception as e:
logging.exception("error: {0}".format(e))
def currentHost(self):
return self.dialog.comboBox.currentText()
@catchExept
def init(self):
for child in self.zk.get_children("/"):
self.treeWidget.addTopLevelItem(
QTreeWidgetItem([child, "/" + child, child]))
@catchExept
def my_listener(self, state):
if state == KazooState.LOST:
# Register somewhere that the session was lost
self.print("state is LOST!")
elif state == KazooState.SUSPENDED:
# Handle being disconnected from Zookeeper
self.print("state is SUSPENDED!")
else:
# Handle being connected/reconnected to Zookeeper
self.print("state is CONNECTED!")
@catchExept
def print(self, text):
logging.debug(text)
@catchExept
def printAllChildren(self, curPath, children, layer):
spaces = " " * layer
for child in children:
newPath = curPath + "/" + child
data, stat = self.zk.get(newPath)
self.print("%s: %s" % (spaces + child, data))
self.printAllChildren(newPath, self.zk.get_children(newPath),
layer + 1)
@catchExept
@pyqtSlot(int)
def closeTab(self, idx):
self.tabWidget.removeTab(idx)
@catchExept
@pyqtSlot(QTreeWidgetItem, int)
def itemOpen(self, item, column):
if not self.zk.exists(item.text(1)):
return
tabName = item.text(2)
for i in range(self.tabWidget.count()):
if tabName == self.tabWidget.tabText(i):
self.tabWidget.setCurrentIndex(i)
return
innerText = QPlainTextEdit()
innerText.setReadOnly(True)
data, stat = self.zk.get(item.text(1))
innerText.setPlainText(data.decode("utf8"))
pos = self.tabWidget.addTab(innerText, tabName)
self.tabWidget.setCurrentIndex(pos)
@catchExept
def drawAllTree(self):
if self.zk.exists("/"):
root = self.zk.get_children("/")
self.printAllChildren("/", root, 0)
else:
self.print("Really?.. How?.. Why?..")
@catchExept
@pyqtSlot(QTreeWidgetItem, int)
def itemClicked(self, item, column):
item.setText(
0,
item.text(2) + " (%s)" % self.getCurrentStat().children_count)
children = item.takeChildren()
newChildren = []
if self.zk.exists(item.text(1)):
for child in self.zk.get_children(item.text(1)):
for oldChild in children:
if oldChild.text(2) == child:
newChildren.append(oldChild)
break
else:
newChildren.append(
QTreeWidgetItem(
[child, item.text(1) + "/" + child, child]))
item.addChildren(newChildren)
